Most of what I build starts from how the work is already done. These are the four principles I keep coming back to, and the tools on this site where each one shows up.
A tool stays in use only if keeping it up to date is an automation's job, or part of work I am already doing. If it needs someone to sit down and feed it every day, it stops being used, however good it is.
My notes are written by agents while I do other work, and get 20 to 60 updates a day.
A rule that someone has to remember will be missed eventually. A rule that code checks runs every time. When something goes wrong once, I write the check that stops it happening again.
My agent setup runs hooks that check every tool call against my rules before it runs. The system architecture model is kept as text in git, so changes to it are reviewed the same way as code.
Teams already have a document control system, email, registers and a browser they are signed in to. I build tools that work in those systems, so nobody has to move their work somewhere new.
The agent connectors read and update records in the systems where they are kept. Shroom works in websites through the Chrome profiles I am already signed in to. Redline opens the PDF and DWG files that arrive for review.
Every fact has one place where it is kept, and everything else points to it. If a person's role changes, one line in one file changes.
Dashboards, reports and the forecasting app read from the same data model. When a number looks wrong, I check it against the source system, not a copy of it.
Everything on my computer is set up in code, so the terminal is how I use it. I tell an AI coding agent what I want, and it works through the apps and systems below.